The df certainly uses a lot of memory, changing dtypes wherever possible and removing columns that are inconsistent or not useful for us
It can also be seen that columns[4:9] contains a lot of null values

Our data consists of months 1-8 (i.e, January to August) for the year 2017.
From above we can see that, compared to the other months, higher amount of transactions were made between 4 & 5 (i.e, April and May).

It seems like a large number of visitors, visit the site during Night time or Late night (i.e, from 9pm - 4am)

Surprising, as even though most of the store visits were done during the night/latenight slot, as we can see above most of the transactions (as well as higher amount transactions) are made during Evening.
Though, the transactions made during night/latenight are on the higher side as well, the transactional amounts are low compared to evening and afternoon.

I assumed the reason why transactions during the night/late night were low even though having the most visit count rate was because the visitors exited the site without triggering any sessions. (i.e, only a single page was visited)

Although, that does not seem the case.
Along with this it can be seen that Evenings have an overall lower bounce count.

Surprising, the transactions made and pageviews are correlated, hence will we put pageviews in our model to see if visitor will transact.
Similarly, source & medium are correlated the most. This is because medium is grouping for source and channel is a grouping for source & medium both combined, yet the correlation factor is less for channel & medium as compared to channel & source.

From the first graph it can be seen that the site is being visited the most through Organic Search and that most of the channels are accessed from a desktop.

Although, in the second graph there is an unexpected finding that even though the visit count is less for Display, through any device, the transactions made/amount is most when the traffic comes from Display channel and are accessed from desktop.

We will take the first 7 and let's check the accuracy

One noticable thing, which I assume is prominent, is that as the pageviews decrease (lower than 8) The sample gets purer hence suggesting that the visitor's chances of transacting is low.
On the other hand as the pageviews increase (around 23 - 25) the sample impurity rises, hence suggesting that the chances of visitors transacting, rises.